A finite population model analysis of co-evolution with matching pennies

Finite Markov models of the evolution of finite populations can be used as a tool to study the theoretical limit behavior of genetic algorithms. Morespecifically, these models can yield exact expectations of performance of genetic algorithms for small optimization problems. In this paper, combinations of these Markov models are used to study the exact expected performance of co-evolutionary genetic algorithms engaged in the Matching Pennies game. We show how changes in the rate of mutation, population size and selection pressure influence the performance of competing genetic algorithms. Download PDF File (0.45MB